Explainable machine learning driven strength degradation investigation of BFRP bar in seawater and sea sand concrete environment

This study applies machine learning (ML) methods to predict the residual tensile strength of Basalt Fiber Reinforced Polymer (BFRP) bars in seawater and sea sand concrete (SWSSC) environment. A total of 384 experimental results of BFRP samples under different corrosive environments are collected to train and test six supervised ML models, including Decision Tree (DT), Random Forest (RF), Extremely Randomized Trees (ERT), Adaptive Boosting (AdaBoost), eXtreme Gradient Boosting Trees (XGBoost), and Artificial Neural Network (ANN). Five evaluation metrics are utilized to compare the performance of ML models and empirical formulae. Comparison analysis results show that ANN, ERT, RF and XGBoost demonstrated excellent predictive capacities, and the accuracy of ML models is superior to that of empirical models. Moreover, based on the SHapley Additive Explanations (SHAP) algorithm, the prediction process of ML models is explained and the impact of variables on the residual capacity of BFRP bars is analyzed. The results indicate that the initial tensile stress has a critical influence on exposure time and temperature exerting significant influences. In contrast, the influence of diameter and tensile rate are negligible. Besides, to mitigate the capacity degradation of BFRP bars, it is suggested that to maintain the environmental temperature below 48℃, ensure that the pH remains within the limit of 12.5, restrict the sustained load to no more than 20 %, and provide concrete protection with its layer thickness exceeds 32 mm. © 2025